6(101' y

Neural Darwinism
A GLOBAL BRAIN

THE ORY

here is one simple principie that governs how the brain works: it evolved; that is, it was not designed. As stated, this principie sounds almost simple-minded, doesn't it? But we must not forget that, although evolution is not intelligent, it is enormously powerful. The power comes from natural selection acting in complex environments over eons of time. A key idea developed by Darwin is embedded in his notion of population thinking: functioning structures and whole organisms emerge as a result of selection among the diverse variant

individuals in a population, which compete with one another for survival. I hold this notion to be central, not only in considering how the brain has evolved, but also in thinking about how it develops and functions. Applying population thinking to understanding how the brain works leads to a global theory, called neural Darwinism or the theory of neuronal group selection. What do we mean by the term "global" and why do we need a global brain theory? An explanation of consciousness will necessarily require an understanding of perception, memory, action, and intentionin short, an overall understanding of how the brain works that goes beyond the functioning of one brain region or another. Given the richness, variety, and range of conscious experience, it is also important to construct a brain theory that is principled and compatible with evolution and development. By principled, I mean a theory that describes the principies governing the major mechanisms by which the brain deals with information and novelty. One such theory or model is the idea that the brain is like a computer or Turing machine. In contrast to such an instructive model, which relies on programs and algorithms, models based on population thinking rely on selection of particular elements or states from a large repertoire of variant elements or states. Explanations of consciousness based on one or the other of these two kinds of models differ greatly. By now, it should be no mystery that I prefer selectional models based on population thinking.

NEURAL DARWINISM

32

33

The reason population thinking is important in determining how the brain works has to do with the extraordinary amount of variation in each individual brain. This is true at all levels of structure and function. Different individuals have different genetic influences, different epigenetic sequences, different bodily responses, and different histories in varying environments. The result is enormous variation at the levels of neuronal chemistry, network structure, synaptic strengths, temporal properties, memories, and motivational patterns governed by value systems. In the end, there are obvious differences from person to person in the contents and styles of their streams of consciousness. The variability of individual nervous systems was commented on by the distinguished neuroscientist Karl Lashley, who admitted that he had no ready explanation for the existence of so much variation. Even though there are general patterns exhibited by the brain in the face of this variation, it cannot be dismissed as mere noise. There is too much of it, and it exists at too many levels of organizationmolecules, cells, and circuits. It is simply not likely that evolution, like a computer programmer dealing with noise, could have devised multiple error-correcting codes to assure preservation of patterns in the brain by counteracting this enormous variation. An alternative way of confronting neural variability is to consider it fundamental and to assume that the individual local differences within each brain make up populations of variants. In this case, selection from such

a population of variants could lead to patterns even under unpredictable circumstances, provided that some constraint of value or fitness was satisfied. In evolution, fitter individuals survive and have more progeny. In the individual brain, those synaptic populations that match value systems or rewards are more likely to survive or contribute more to the production of future behavior. This view is in sharp contrast to computer models of the brain and mind. According to these models, signals from the environment carry input information that is unambiguous, once contaminating noise is averaged away or otherwise dealt with. These models assume that the brain has a set of programs, or so-called effective procedures, which are capable of changing states based on the information carried by the inputs, yielding functionally appropriate outputs. Such models are instructive in the sense that information from the world is assumed to elicit the formation of appropriate responses based on logical deduction. These models do not deal, however, with the fact that inputs to the brain are not unambiguousthe world is not like a piece of tape with a fixed sequence of symbols for the brain to read. I have already mentioned the challenge to computer models of the brain posed by the richly variable circuitry of real brains. There is also a set of functional issues that make computer models unlikely. For example, the mapped connections from the sense of touch in the hand through the thalamus to the region of somatosensory cortex are variable and plastic, even in adults. The sub-

NEURAL DARWINISM

NEURAL DARWINISM

34

35

regions in the somatosensory cortex mapping the fingers dynamically shift all their boundaries as a result of excessive use of even one fingera shift in the context of use. Similar phenomena reflecting such context dependence and dynamic circuit variation are seen for other senses. Furthermore, in sensory systems such as that for vision, there are multiple cortical regions that are each functionally segregated, for example, for color, movement, orientation, and so on. These functionally specialized areas can exceed thirty in number and are distributed afl over the brain. Yet there is no superordinate area or executive program binding the color, edge, form, and movement of an object into a coherent percept. This binding is not explicable by invoking a visual computer program operating according to the principies of artificial intelligence. A coherent percept in fact nevertheless emerges in various contexts, and explaining how this occurs constitutes the so-called binding problem. A global brain theory must provide a cogent solution to this problem by proposing an appropriate mechanism. It will soon become clear that such a solution is central to our understanding of consciousness. To emphasize the dependence of perception on context, we may call upon the huge phenomenology of illusions, visual and otherwise. One example is the Kanizsa pattern, which consists of the angular portions of a triangle, disconnected, but appears to show an overlying triangle with sharp boundaries (Figure 4). Yet there is no true energy difference in the light that is

Figure 4. Illusory contours in a Kanizsa triangle. Most people report the appearance of a distinct triangular shape and an increase in apparent luminance within the triangle, but neither of there features exists in the physical image. received from the two sides of the contour that is perceived. Such a contour is called "illusory." The brain constructs the contour, which, by the way, is not necessarily a straight line but can be curved depending on the context of the particular figure used. Many other functional responses of the perceiving animal could be described to illustrate why an a priori program is not a likely explanation for physiological or psychological properties. I shall mention only two more. The first is the remarkable tendency of brains to seek out closure and avoid gaps. In daily life, for example, you do not see the blind spot in your visual field occasioned by the presence of the optic nerve near the center of your retina. Even more striking phenomena come

NEURAL DARWINISM

36

NEURAL DARWINISM

37

from the field of neuropsychology, which, among other things, studies responses to strokes. This field is replete with examples of closure phenomena that can even be delusional. A most exotic example is anosognosia, a syndrome in which a paralyzed patient does not recognize the existence of paralysis even if it involves his or her entire left side. In such cases, we see extraordinary adaptation and integration by the damaged brain as it responds to the loss of cortical areas. In addition to construction and closure, and possibly in connection with them, the brain's capacity to generalize is astonishing. A case in point is the ability of pigeons, when appropriately rewarded, to look at numerous photographs of various fish species in different scales and contexts and learn to positively recognize the similarity in the photographs. Pigeons trained at this task can recognize that these diverse pictures have something in common more than 80 percent of the time. It is highly unlikely that this behavior is the result of a fixed template or a set of predetermined algorithms in the brains of pigeons. Nor can it be explained by natural selection for the positive recognition of fish. Pigeons neither evolve with fish nor live with them, and they don't eat them either. I could cite many more examples ranging from the developmental anatomy of the brain to the individual variation of brain scans in humans carrying out similar tasks. But the conclusion is clear: the brains of higherlevel animals autonomously construct patterned re-

sponses to environments that are full of novelty. They do not do this the way a computer doesusing formal rules governed by explicit, unambiguous instructions or input signals. Once more, with feeling: the brain is not a computer, and the world is not a piece of tape. If the brain is in fact not a computer and the world is not a piece of tape, how can the brain operate so as to yield adaptive and patterned responses? As I have already suggested, the answer lies in a selectionist theory that I have called the theory of neuronal group selection, or TNGS (Figure 5). This theory has three tenets: (1) Developmental selectionduring the early establishment of neuroanatomy, epigenetic variations in the patterns of connections among growing neurons create repertoires in each brain area consisting of millions of variant circuits or neuronal groups. The variations arise at the level of synapses as a result of the fact that neurons that fire together wire together during the embryonic and fetal stages of development. (2) Experiential selectionoverlapping this first phase of selection and alter the major neuroanatomy is built, large variations in synaptic strengths, positive and negative, result from variations in environmental input during behavior. These synaptic modifications are subject to the constraints of value systems described in the previous chapter. (3) Reentryduring development, large numbers of reciprocal connections are established both locally and over long distances. This provides a basis for signaling between mapped areas across such reciprocal fibers. Reentry is

NEURAL DARWINISM

38

NEURAL DARWINISM

39

Developmental selection (Yielding primary repertoire)

Cell division Cell death Process extension

49/

Experiential selection (Yielding secondary repertoire)

Changes in strength of populations o synapses

Reentrant mapping

Input to Map 1

Input to Map 2

Input to Map 1

Input to Map 2

Figure 5. The three main tenets of the theory of neuronal group selection, or neural Darwinism: (1) Developmental selection leads to a highly diverse set of circuits, one of which is shown. (2) Experiential selection leads to changes in the connection strengths of synapses, favoring some pathways (thickened black Enes) and weakening others (dashed unes). (3) Reentrant mapping, in which brain maps are coordinated in space and time through ongoing reentrant signaling across reciprocal connections. The black dots in the maps on the right indicate strengthened synapses. As a result of (1) and (2), a myriad of circuits and functioning pathways is created constituting a repertoire for selectional events. The further and ongoing events of reentry in (3) must be thought of as dynamic and recursive, mapping the maps over time.

the ongoing recursive interchange of parallel signals among brain areas, which serves to coordinate the activities of different brain areas in space and time. Unlike feedback, reentry is not a sequential transmission of an error signal in a simple loop. Instead, it simultaneously involves many parallel reciprocal paths and has no prescribed error function attached to it. The consequence of this dynamic process is the widespread synchronization of the activity of widely distributed neuronal groups. It binds their functionally segregated activities into circuits capable of coherent output. In the absence of logic (the organizing principie of computers as instructive systems), reentry is the central organizing principie that governs the spatiotemporal coordination among multiple selectional networks of the brain. This solves the binding problem that I mentioned earlier. Through reentry, for example, the color, orientation, and movement of a visual object can be integrated. No superordinate map is necessary to coordinate and bind the activities of the various individual maps that are functionally segregated for each of these attributes. Instead, they coordinate by communicating directly with each other, through reentry. The three tenets of the TNGS together form a selectional system. Prominent examples of selectional systems include evolution, the immune system, and complex nervous systems. All follow a set of three guiding principies. The first principle assumes a means for generating diversity in a population of elements, whether

NEURAL DARWINISM 41

of individuals or of cells. The second is a means allowing extensive encounters between individuals in a variant population or repertoire and the system that is to be recognized, whether it is an ecological environment, a foreign molecule, or a set of sensory signals. The third principie is some means to differentially amplify the number, survival, or influence of those elements in the diverse repertoire that happen to meet selective criteria. In evolution, these are criteria of fitness allowing the differential survival and breeding of certain individualsthe process of natural selection itself. In immunity, amplification occurs through the enhanced division of just those clones of immune cells having antibodies on their surface that bind particular foreign molecules or antigens weli enough to exceed a certain critica' energy of binding. In neural systems, amplification consists of enhancing the strengths of those synapses and circuits of neuronal groups that meet the criteria set by value systems. It is the neuronal groups made up of excitatory and inhibitory neurons in particular anatomical patterns rather than individual neurons that are selected. Notice that while these three different selectional systems obey similar principies, they use different mechanisms to achieve successful matching to various unforeseen inputs. Evolution is, of course, special and overarching because it is also responsible for actually selecting the different mechanisms used by the immune and nervous systems. It tends to favor those individuals that successfully utilize such mechanisms to improve

their fitness and allow more of their progeny to survive. Since the proposal of the TNGS in 1978, a growing body of evidence has supported the notion that neuronal groups connected by reentrant interactions are the selectional units in higher-level brains. This evidence is presented in a number of books and papers and will not be reviewed here. Instead, I will consider certain consequences of the theory that are particularly important for understanding the mechanisms underlying consciousness. One important consequence is that the brain is so versatile in its responses because those responses are degenerate. Degeneracy is the ability of structurally different elements of a system to perform the same function or yield the same output. A clear-cut example is seen in the genetic code. The code is made up of triplets of nucleotide bases, of which there are four kinds: G, C, A, and T. Each triplet, or codon, specifies one of the twenty different amino acids that make up a protein. Since there are sixty-four different possible codonsactually sixty-one, if we leave out three stop codonswhich makes a total of more than one per amino acid, the code words are degenerate. For example, the third position of many triplet codons can contain any one of the four letters or bases without changing their coding specificity. If it takes a sequence of three hundred codons to specify a sequence of one hundred amino acids in a protein, then a large number of different base se-

NEURAL DARWINISM

NEURAL DARWINISM

42

43

quences in messages (approximately 3100) can specify the same amino-acid sequence. Despite their different structures at the level of nucleotides, these degenerate messages yield the same protein. Degeneracy is a ubiquitous biological property. It requires a certain degree of complexity, not only at the genetic level as I have illustrated aboye, but also at cellular, organismal, and population levels. Indeed, degeneracy is necessary for natural selection to operate and it is a central feature of immune responses. Even identical twins who have similar immune responses to a foreign agent, for example, do not generally use identical combinations of antibodies to react to that agent. This is because there are many structurally different antibodies with similar specificities that can be selected in the immune response to a given foreign molecule. Degeneracy is particularly important in helping to solve major problems in complex nervous systems. I have already mentioned the binding problem. How can it be that, despite the absence of a computer program, executive function, or superordinate map, up to thirtythree functionally segregated and widely distributed visual maps in the brain can nevertheless yield perception that coherently binds edges, orientations, colors, and movement into one perceptual image? How do different maps for color, orientation, object movement, and so on correlate or coordinate their responses? As I suggested aboye, the answer lies in mutual reentrant interactions that, for a time, link various neuronal groups in each

map to those of others to form a functioning circuit. Simulations show that the neurons that yield such circuits fire more or less in phase with each other, or synchronously. But in the next time period, different neurons and neuronal groups may form a structurally different circuit, which nevertheless has the same output. And again, in the succeeding time period, a new circuit is formed using some of the same neurons, as well as completely new ones in different groups. These different circuits are degeneratethey are different in structure but they yield similar outputs to solve the binding problem (Figure 6). Within each particular circuit, the different neuronal groups fire synchronously. The different circuits yielding the same output are not, however, synchronous or in phase with each other, nor do they have to be. As a result of reentry, the properties of synchrony and coherency allow more than one structure to give a similar output. As long as such degenerate operations occur in succession to link distributed populations of neuronal groups, there is no need for an executive or superordinate program as there would be in a computer. The formulation of a global brain theory like the TNGS, while essential to understanding how the brain works, does not solve all of the detailed mechanistic problems related to the local operations of networks in the various nuclei and regions of the brain. But it does remove the paradoxes that arise if one assumes that the brain functions like a computer. One such paradox

NEURAL DARWINISM

NEURAL DARWINISM

44

45

OUTPUT

munculus, a little man who lives in the brain, to interpret the meaning of a percept. Just as Darwin's theory of natural selection disposed of the argument from design, the TNGS disposes of the need for either a fixed instructional plan or a homunculus in the head. These issues are directly relevant to my next task, which is to show how the principies and mechanisms of the TNGS can be used to understand the origin of consciousness.

Outputs:

At time t, C At time t+1, A At time t+2, B

Figure 6. Illustration of the degeneracy of reentrant circuits in the brain. Even though the three overlapping circuits in A, B, and C are different, as shown by the shading, they can yield a similar output over some period of time. would have us imagine a cell with a designated categorical function that dominates the function of all subordnate neurons connected to itfor example, a cell that fires when you think of a particular person, a so-called grandmother cell. Such a cell is not necessary in this theory. Different cells can carry out the same function and the same cell can, at two different times, carry out different functions in different neuronal groups. Moreover, given the selectional nature of higher-order interactions in the brain, one does not have to invoke a ho-

NEURAL DARWINISM

NEURAL DARWINISM

46

47